EP 01 · 5 min read

Why Machine-Learning Models Struggle to Separate True Positives from False Positives

A model can look convincing in a dashboard and still make poor decisions. One reason is that a model score is not the same thing as a correct prediction. Another is that the examples a model calls...

EP 02 · 4 min read

Why Analysing the Entire Dataset Is Different from Analysing Model-Selected Predictions

A feature can look useful across an entire dataset and become much less useful among the rows a model actually selects. The reverse can also happen: a feature with a weak overall relationship to the...

EP 03 · 4 min read

ROC-AUC, Precision and Recall: Understanding What Your Model Is Actually Telling You

Model evaluation can become confusing when several metrics move in different directions. A classifier may have a respectable ROC-AUC but disappoint when converted into a list of selected candidates....

EP 04 · 4 min read

The Danger of Drawing Conclusions from Small Samples

A result can be mathematically correct and still be a weak basis for a decision. If a model selects five cases and all five are positive, the observed precision is 100%. But five observations do not...

EP 05 · 3 min read

Understanding Feature Distributions, Outliers and Extreme Values

Machine-learning models do not receive the world as it is. They receive columns of numbers created by data pipelines. Those numbers may be skewed, sparse, rounded, duplicated, truncated or affected...

EP 06 · 4 min read

How a Small Denominator Creates Misleading Growth Ratios

Ratios are attractive features because they compress two measurements into one number. A growth ratio can describe how much activity increased relative to an earlier baseline. But the ratio can...

EP 07 · 3 min read

Cumulative Volume vs Incremental Volume: Understanding Trading Activity

Two features can sound as though they measure the same thing while describing different behaviour. Cumulative volume tells us how much activity has occurred since the beginning of a window....

EP 08 · 4 min read

Multicollinearity: When Multiple Features Measure Almost the Same Thing

A dataset can contain dozens of features and still describe only a handful of underlying behaviours. Several columns may be different versions of the same measurement: cumulative volume over adjacent...

EP 09 · 3 min read

Why Feature Interactions Can Matter More Than Individual Features

A feature can be useful only under certain conditions. High volume may mean something different when liquidity is thin than when liquidity is deep. A large growth ratio may be more informative when...

EP 10 · 4 min read

Measuring Growth Quality Rather Than Growth Magnitude

Growth magnitude asks how large an increase was. Growth quality asks what kind of activity produced it. In a fast-moving dataset, an impressive number can hide a weak foundation: a tiny starting...

EP 11 · 3 min read

Using Quartiles to Discover Nonlinear Relationships

A feature does not always have a simple relationship with an outcome. More volume may be associated with better outcomes up to a point, after which the relationship flattens. A buyer-count feature...

EP 12 · 3 min read

Feature Selection and Ablation: Why More Features Don't Always Mean Better Models

Adding another feature can make a model look more sophisticated, but it does not guarantee better predictions. A new column may duplicate information already present, introduce noise, create leakage...

EP 13 · 3 min read

Logistic Regression vs Gradient Boosting: Which Is Better for Predictive Trading?

When comparing algorithms, it is tempting to ask which one is best. A more useful question is: which model is better suited to this dataset, this validation design and this decision? Logistic...

EP 14 · 3 min read

Meta-Labelling: Using a Second Model to Evaluate the First

A first-stage model can be useful at finding a broad set of candidates without being precise enough for a narrower decision. Meta-labelling adds a second model that evaluates the cases selected by...

EP 15 · 3 min read

Regularisation: Preventing Machine-Learning Models from Learning Noise

A model can fit historical data extremely well and still perform poorly on new observations. This happens when the model learns patterns that are specific to the training sample rather than...

EP 16 · 3 min read

Testing New Features Without Changing Your Existing Model Architecture

When a model disappoints, the instinct is often to change several things at once: replace the algorithm, redesign the architecture, tune dozens of parameters and add a new feature set. That makes it...

EP 17 · 4 min read

Out-of-Fold Predictions: How to Evaluate Models Honestly

A model's predictions on data it has already trained on are not a fair estimate of how it will behave on unseen observations. Training data has influenced the fitted model, so its scores can be...

EP 18 · 3 min read

Chronological Cross-Validation and Purging: Preventing Data Leakage

Randomly splitting a time-dependent dataset can make a model appear more capable than it really is. If a model trains on later observations and is evaluated on earlier ones, information about future...

EP 19 · 3 min read

Choosing the Right Prediction Threshold: Precision vs Recall

A classifier may produce a probability score for every observation. To turn those scores into a selected list, someone must choose a threshold. A score above the threshold is classified as positive;...

EP 20 · 4 min read

From Prediction to Profitability: Expectancy, Sharpe Ratio and Drawdown

A model can classify historical outcomes reasonably well and still fail to produce a viable real-world process. Prediction metrics describe how well a model matches labels. Profitability metrics...

SOLANA · 3 min read

What really happens in the first hour of a pump.fun launch

Two months of pump.fun launches with a real first 15 minutes: how often they reach 2x, 5x and 10x, how front-loaded volume is, and when peaks happen.